Research

Optimizing AKS costs starts with visibility across clusters, namespaces, applications, owners, and resources. Combine HPA, KEDA, VPA, and Cluster Autoscaler with clear safeguards. Use spot nodes for interruption-tolerant workloads, rightsize CPU and memory based on historical data, and adjust node pools accordingly. Sustain savings through ownership labels, budgets, anomaly alerts, showback, and regular FinOps reviews.

This article concludes a series about cloud security in existing Azure environments. Together, they reveal something more important: cloud security does not work as a collection of controls. It only works as a system.

I’ve explained that I implemented Azure Cosmos DB for storing save-games for a private project. As a software developer, I’ve found the technology Cosmos DB to be deceptively tricky to grasp, and hopefully my last blog has helped you wrap your head around designing your database in terms of partitioning and scaling. This time I’ll dive deep into securing the database and how to make data available at lightning-fast speeds all over the world!

This article is the fifth part in a series about cloud security within existing Azure environments. This section moves away from technical details and addresses a more challenging topic: ownership. Most cloud security issues arise not from absent controls but from unclear responsibilities.

I still remember participating in “failover tests” years ago (decades, actually — but let’s not dwell on that) for a financial institution, based on restoring data from tape. This was a company with over 300 physical servers, relying on time-consuming recovery from a sequential medium to bring back systems like Swift and large Oracle databases at a provider’s disaster recovery site.

Part 2: The inner workings of Retrieval Augmented Generation (RAG)